SimBIID (version 0.2.2)

ABCRef: Produces ABC reference table

Description

Produces reference table of simulated outcomes for use in various Approximate Bayesian Computation (ABC) algorithms.

Usage

ABCRef(
  npart,
  priors,
  pars,
  func,
  sumNames,
  parallel = FALSE,
  mc.cores = NA,
  ...
)

Value

An data.frame object with npart rows, where the first p columns correspond to the proposed parameters, and the remaining columns correspond to the simulated outputs.

Arguments

npart

The number of particles (must be a positive integer).

priors

A data.frame containing columns parnames, dist, p1 and p2, with number of rows equal to the number of parameters. The column parname simply gives names to each parameter for plotting and summarising. Each entry in the dist column must contain one of c("unif", "norm", "gamma"), and the corresponding p1 and p2 entries relate to the hyperparameters (lower and upper bounds in the uniform case; mean and standard deviation in the normal case; and shape and rate in the gamma case).

pars

A named vector or matrix of parameters to use for the simulations. If pars is a vector then this is repeated `npart` times, else it must be a matrix with `npart` rows. You cannot specify both `pars` and `priors`.

func

Function that runs the simulator. The first argument must be pars. The function must return a vector of simulated summary measures, or a missing value (NA) if there is an error. The output from the function must be a vector with length equal to length(sumNames).

sumNames

A character vector of summary statistic names.

parallel

A logical determining whether to use parallel processing or not.

mc.cores

Number of cores to use if using parallel processing.

...

Extra arguments to be passed to func.

Details

Runs simulations for a large number of particles, either pre-specified or sampled from the a set of given prior distributions. Returns a table of summary statistics for each particle. Useful for deciding on initial tolerances during an ABCSMC run, or for producing a reference table to use in e.g. the ABC with Random Forests approach of Raynal et al. (2017).

References

Raynal, L, Marin J-M, Pudlo P, Ribatet M, Robert CP and Estoup A. (2017) <ArXiv:1605.05537>

Examples

# \donttest{
## set up SIR simulation model
transitions <- c(
    "S -> beta * S * I -> I", 
    "I -> gamma * I -> R"
)
compartments <- c("S", "I", "R")
pars <- c("beta", "gamma")
model <- mparseRcpp(
    transitions = transitions, 
    compartments = compartments,
    pars = pars
)
model <- compileRcpp(model)

## generate function to run simulators
## and produce final epidemic size and time
## summary statistics
simRef <- function(pars, model) {
    ## run model over a 100 day period with
    ## one initial infective in a population
    ## of 120 individuals
    sims <- model(pars, 0, 100, c(119, 1, 0))
    
    ## return vector of summary statistics
    c(finaltime = sims[2], finalsize = sims[5])
}

## set priors
priors <- data.frame(
    parnames = c("beta", "gamma"), 
    dist = rep("gamma", 2), 
    stringsAsFactors = FALSE
)
priors$p1 <- c(10, 10)
priors$p2 <- c(10^4, 10^2)

## produce reference table by sampling from priors
## (add additional arguments to 'func' at the end)
refTable <- ABCRef(
    npart = 100, 
    priors = priors, 
    func = simRef, 
    sumNames = c("finaltime", "finalsize"),
    model = model
)
refTable
# }